CRITICAL PERIOD PLASTICITY IN LOCAL CORTICAL CIRCUITS

Transcription

1 CRITICAL PERIOD PLASTICITY IN LOCAL CORTICAL CIRCUITS Takao K. Hensch Abstract Neuronal circuits in the brain are shaped by experience during critical periods in early postnatal life. In the primary visual cortex, this activity-dependent development is triggered by the functional maturation of local inhibitory connections and driven by a specific, late-developing subset of interneurons. Ultimately, the structural consolidation of competing sensory inputs is mediated by a proteolytic reorganization of the extracellular matrix that occurs only during the critical period. The reactivation of this process, and subsequent recovery of function in conditions such as amblyopia, can now be studied with realistic circuit models that might generalize across systems. CRITICAL PERIOD A strict time window during which experience provides information that is essential for normal development and permanently alters performance. SENSITIVE PERIOD A limited time during development, during which the effect of experience on brain function is particularly strong. AMBLYOPIA Poor vision through an eye that is otherwise physically healthy due to little or no transmission of the visual image to the brain through circuits that are hardwired during a developmental critical period. It affects 2 5% of the population. RIKEN Brain Science Institute, 2-1 Hirosawa, Wako-shi, Saitama, , Japan. hensch@riken.jp doi: /nrn1787 From polyglots 1 to virtuosi 2,3, human performance reflects the neural circuits that are laid down by early experience. Although learning is possible throughout life, there is no doubt that those who start younger fare better, and that plasticity is enhanced during specific windows of opportunity. An understanding of the neural basis of such CRITICAL or SENSITIVE PERIODS of brain development would inform not only classroom and educational policy, but also drug design, clinical therapy and strategies for improved learning into adulthood. Although which might be the critical periods for higher cognitive functions such as language, music or emotional control is the subject of popular debate, such sweeping questions fail to acknowledge the sequential nature of a multistage process that involves many brain regions. Evidence about the acquisition of these rich human behaviours is largely anecdotal, as the complex environments of everyday life are inherently difficult to quantify and normalize across individuals. Clear evidence for critical periods has instead been found in the primary sensory systems of several species 4. Animal models are now revealing, with greater resolution, the molecular, cellular and structural events that underlie experience-dependent circuit refinement. In this review, I focus on the primary visual cortex, which has been the premier model of critical period plasticity for 40 years 5. During a brief postnatal period (of weeks to years, in proportion to the expected lifespan of the species 6,7 ), the closure of one eye (but not both) causes a permanent loss of visual acuity through that eye. AMBLYOPIA occurs despite there being no damage to the retina or its target, the visual thalamus (dorsal lateral geniculate nucleus, dlgn) and is determined in the neocortex (in the visual primary cortex, V1), where the inputs from the two eyes first converge and compete for space 5,6. Most impressively, the seemingly innocuous act of covering an eye can profoundly alter the physical structure of the brain during the critical period only. Shaping column size during the critical period The cortical column is a fundamental unit of organization of the mammalian neocortex. Clusters of thalamocortical axon terminals that serve either the right or left eye tesselate layer 4 of the mature cortex to produce alternating OCULAR DOMINANCE domains 8,9. Occluding one eye during development yields an expansion of the columns serving the open eye at the expense of those responding to the deprived eye, which become reduced in size and afferent complexity 10,11. This physical manifestation of early postnatal experience occurs gradually, and is preceded by more rapid changes 12,13 of intra cortical circuits outside layer 4, which instruct the hard-wiring of an anatomical fingerprint that is unique to the individual. By extension, the segregation of columns by normal vision during the critical period has NATURE REVIEWS NEUROSCIENCE VOLUME 6 NOVEMBER

2 Mature columns Activation profile Local cortical circuits Developing thalamocortical inputs Left eye Right eye OCULAR DOMINANCE Relative anatomical or physiological strength of connections from either eye to individual cells in the primary visual cortex. THALAMOCORTICAL AFFERENTS Axons from the thalamus (for example, the dlgn) that relay sensory input from the periphery (for example, the retina) to layer 4 of the neocortex. BENZODIAZEPINES Modulate chloride flux through GABA A receptors that contain the γ2 and any combination of α1, α2, α3 or α5 subunits. Benzodiazepine agonists enhance and inverse agonists decrease GABA efficacy. Thin Normal Wide Lateral inhibition (DMCM) + Lateral inhibition (diazepam) Figure 1 Local circuit control of developing columnar architecture in the neocortex. Activity-dependent models of segregation predict that cortical GABA (γ-aminobutyric acid) circuits are involved in determining the final column spacing from an initially overlapping mosaic 17,18. Neuronal activity from thalamic inputs serving the right or left eye is spread within the nearby cortex by local excitatory connections (red cell) but is inhibited at greater distances (blue cell), producing a mexican hat activation profile. When this profile is modulated during development by preferentially enhancing or reducing horizontal, long-range inhibition (arrows), columns emerge that are wider or thinner than normal, respectively. This hypothesis was verified in vivo in kittens by modulating GABA A (GABA type A) currents 20 with benzodiazepine agonists (diazepam) or inverse agonists (methyl-6,7-dimethoxy-4-ethyl-β-carboline, DMCM) throughout the critical period 19. been thought to result from similar activity-dependent rules acting on an initially overlapping continuum of THALAMOCORTICAL AFFERENTS. This dogma has recently been challenged by the finding that axon terminals of thalamocortical afferents might, in part, be clustered well before the critical period 14. There is also some evidence that siblings show substantial similarity of visual maps, which supports the idea that molecular cues establish columnar architecture 15. Nonetheless, it is agreed that sensory experience is important for individualizing ocular dominance maps during the critical period. Even the focal patterns of deprivation that are produced by shadows of blood vessels in a single eye are embossed onto the primary visual cortex 16. In computational models of self-organization, the recipient cortical circuits largely determine the final spacing of columns 17,18. Overlapping inputs segregate into clusters ( neurons that fire together wire together ) through a neocortical organization that spreads excitation locally but is limited at a distance by farther-reaching inhibition. FIGURE 1 shows how a central area of excitation surrounded by a larger concentric area of inhibition (the canonical mexican hat profile) of intracortical activation could, in theory, influence thalamic innervation. In particular, lateral inhibition can establish narrow or wide columns in computer simulations of column formation by adjusting the contrast between inputs at nascent border regions during this winner-takes-all competition 18. These long-standing theoretical predictions have recently been validated in vivo through the direct infusion of BENZODIAZEPINES into the kitten visual cortex during the critical period 19. Such drugs come in three varieties, including agonists such as diazepam, inverse agonists such as the β-carbolines (for example, methyl-6,7-dimethoxy-4-ethyl-β-carboline DMCM), and antagonists that block the actions of both (flumazenil) 20. All bind to a particular subset of GABA A (γ-aminobutyric acid type A) receptor, the agonists enhancing and inverse agonists reducing chloride flux. Increasing inhibition throughout the critical period with benzodiazepine agonists leads to a local 30% increase in column width, whereas inverse agonists produce column shrinkage 19 (FIG. 1). Interestingly, increased column spacing has also been reported for STRABISMUS through exotropic deviation of the eyes during the critical period 21. Both enhanced lateral inhibition by direct intracortical infusion of diazepam and decorrelation of visual input by artificial squint are conditions that favour the maximal segregation of ocular dominance. In summary, local imbalances of neuronal activity influence columnar architecture during normal development, and this cannot be explained solely by genetic instruction. The broad range of column sizes observed in nature 22 might reflect individual differences in the span of local circuit inhibition. Detection by excitatory inhibitory balance Long-range, horizontal inhibitory axons therefore provide a scaffold for discriminating competing sensory input (for example, right or left eye) in the developing neocortex. This is in contrast to the maturing neuromuscular junction in the periphery, where individual motor axons may compete directly with each other for synaptic space on the same end plate 23. This competition is limited by the global resources (for example, neurotransmitter) that are available within individual presynaptic arbors, each of which might contact multiple muscle fibres 24. In the tangled circuits of the neocortex 25,26, the action of locally balanced excitation and inhibition first integrates and then detects competition. Even small changes in the relative amounts of excitation and inhibition can markedly alter information processing 27. This delicate balance is dynamically adjusted by circuitry in the cortical layers 28 30, where inhibitory connections are developed later than excitatory connections in the pre-critical period for ocular dominance 31. Drastic pharmacological perturbations of neuronal activity, such as hyperexcitation (by, for example, glutamate 32 or bicuculline 33 ) or total silencing (by, for instance, TETRODOTOXIN (TTX) 34, 2-amino- 5-phosphonovaleric acid (APV) 35 or muscimol 36,37 ), not surprisingly disrupt plasticity, but fail to inform us about intrinsic network behaviour involved in this process. Gentle titration of endogenous neurotransmission by gene-targeted disruption in mice has been instrumental in dissecting the physiological role of 878 NOVEMBER 2005 VOLUME 6

3 a b Cells (%) Ocular dominance Plasticity and loss of visual acuity Enhance GABA function (Benzodiazepines, BDNF overexpression) Cell responses to the contralateral eye Critical period Reduce GABA function (dark-rearing, delete Gad65) Postnatal age Figure 2 GABA-mediated control of the critical period. a Monocular deprivation produces a loss of response to the deprived eye and a gain of open-eye input, as measured by the neuronal discharge of single units from the mouse visual cortex 38. The ocular dominance of cells, rated on a seven-point scale of neuronal responsiveness, indicates a typical bias toward the contralateral eye (1 3) in the rodent (top left). After 3 or more days of monocular deprivation, the distribution shifts toward the open, ipsilateral eye (4 7; top right). b Sensitivity to monocular deprivation is restricted to a critical period that begins, in mice, about 1 week after the eyes open (at postnatal day 13) and peaks 1 month after birth 38. Monocular deprivation causes amblyopia only during the same critical period. Red circles indicate the onset, peak and end of amblyopia resulting from monocular deprivation 39. The onset of plasticity can be delayed by directly preventing the maturation of GABA (γ-aminobutyric acid)-mediated transmission by gene-targeted deletion of Gad65, which encodes a GABA-synthetic enzyme 43, or by dark-rearing from birth (red arrow) Conversely, the critical period can be brought forward by enhancing GABA transmission directly with benzodiazepines just after eye-opening or by promoting the rapid maturation of interneurons through excess brain-derived neurotrophic factor (BDNF) expression (blue arrow) 47,48. STRABISMUS Deviation of the two eyes due to a weakening of extraocular musculature that results in either an inward (esotropic) or outward (exotropic) rotation of one orbit and consequent amblyopia. Cells (%) local circuit elements. As in other species 5 9, MONOCULAR DEPRIVATION (MD) shifts the spiking response of neurons in the mouse V1 towards the open eye 38, but, again, only during a critical period for behavioural amblyopia 39 (FIG. 2a; see also BOX 1). The primary inhibitory neurotransmitter in the brain, GABA (γ-aminobutyric acid), is synthesized by glutamic acid decarboxylase produced by two distinct genes, Gad65 and Gad67. Of these, Gad65 is concentrated in axon terminals and bound to synaptic vesicles whereas Gad67 is found throughout the cell 40. Deletion (knockout) of Gad67 is lethal and eliminates most cortical GABA content 41, but Gad65-knockout mice are viable and show poor GABA release only on strong stimulation 42,43. Baseline receptive field prop erties are normal in the absence of GAD65, but ocular dominance plasticity is prevented until inhibition is acutely restored with diazepam 43. When infused directly into V1, the use-dependent nature and rapid breakdown of benzodiazepine agonists ensures that only those local inhibitory circuits that are normally engaged by monocular deprivation will be boosted. Remarkably, rescue of plasticity is possible at any age in Gad65-knockout mice, which indicates that the critical period is dependent on the proper level of inhibitory transmission (FIG. 2b) 44. Conversely, the onset of the critical period can be accelerated by prematurely enhancing inhibition with benzodiazepines just after eye opening 44 46, as well as by transgenic overexpression of brain-derived neurotrophic factor (BDNF) to promote the maturation of GABA neurons 47,48 (FIG. 2b). A close relationship between neuronal activity, BDNF release and GABA function also explains the classic effect of dark-rearing. Raising animals without visual experience from birth naturally reduces BDNF levels 49 and GABA-mediated transmission in the visual cortex 50,51, and delays the peak of plasticity into adulthood 45,52,53. Direct diazepam infusion 45, or BDNF overexpression 54 or secretion by enriched environments 55 in complete darkness abolish the expected delay of the critical period. These striking results indicate that tonic GABA release is sufficient to trigger the eventual closure of the plastic state, even in the total absence of visual input. Specific GABA circuits for plasticity Interestingly, not all GABA circuits are involved in critical period regulation. Several lines of evidence point towards a single class of interneuron that has the potential to mediate long-range inhibition and synchrony in the visual cortex. Among the many types of GABA-positive interneuron (FIG. 3b), neurochemical markers, such as calcium-binding proteins, have been used to reveal that the onset of the critical period corresponds closely to the emergence of PARVALBUMINpositive cells 59 and both events are accelerated by BDNF overexpression 47. The specific blockade of a potassium channel (Kv3.1) that uniquely regulates the fidelity of FAST-SPIKING behaviour (and thereby GABA release) from parvalbumin-positive interneurons slows the rate of ocular dominance plasticity (Y.-T. Matsuda et al., unpublished observations). To date, the molecular biology of benzodiazepine action has provided the deepest insight into the local circuits that underlie plasticity. The α-subunits of GABA A receptors determine benzodiazepine binding through a single amino acid residue in their amino terminus 20,63 (FIG. 3a). In mice, knock-in of a point mutation at this site renders individual GABA A receptor subtypes insensitive to benzodiazepines 64. Weak inhibition in the visual cortex in early life (as observed for Gad65 deletion) prevents experience-dependent plasticity 44,45. Loss of responsiveness to an eye deprived of vision can be NATURE REVIEWS NEUROSCIENCE VOLUME 6 NOVEMBER

4 Box 1 Critical periods: gateway to lifelong plasticity Plasticity Eye opening TETRODOTOXIN (TTX). A voltage-dependent sodium channel blocker that can be used to silence all neural activity except spontaneous neurotransmitter release. MONOCULAR DEPRIVATION (MD). Imbalanced visual input due to the occlusion of one eye by patching, eyelid suture or intraocular TTX injection. Development Ocular dominance is one of several thalamocortical circuit properties in the primary visual cortex (V1) that develop in an experience-dependent manner during early life 4,186. These might show different postnatal profiles (see panel) with, for example, orientation or direction selectivity (green) being shaped earlier than ocular dominance 7,44,53 (orange), or slow-wave sleep oscillations 169 (dark blue). Surprisingly, the detailed molecular machinery underlying each receptive field property may also differ 53. Recent gene expression profiling supports the idea that the critical period for ocular dominance offers a specialized molecular milieu for plasticity 170,171, which is consistent with dendritic spine and axonal rearrangement being limited to this time in life 10,85,86. Recently, monocular deprivation in adult animals has been reported to produce subthreshold synaptic plasticity , which has no further impact on spiking output 38,44,175 spines 85,86 or visuospatial acuity 7,39. This suggests that the closure of the critical period might reflect sequential locks that are placed on the molecular pathway as it flows from mature GABA (γ-aminobutyric acid)-mediated detection towards structural consolidation (FIG. 6). Functional plasticity may persist throughout life in many other systems (such as the somatosensory cortex 154,176 and the barn owl tectum 149 ). Interestingly, limited plastic states can be greatly facilitated (to near critical period levels) by the judicious activation of neuromodulatory systems in adulthood 177,178. A potent endogenous regulator of neuromodulation is sleep, which plays an intriguing part in enhancing adult learning 179. Neuromodulators signal through second-messenger systems (FIG. 6), which can acutely reset the excitatory inhibitory balance 180,181. Full reactivation of the critical period may further require the undoing of hard-wired neuronal structures 72,73. The non-permissive growth environment of mature myelin is an especially promising target for rekindling plasticity, as it relates broadly to the acquisition of visual 182, linguistic 183 or musical 184 abilities. Better still, laying down multiple anatomical traces through rich childhood experiences might also expand the capacity for plasticity in adulthood 147,150. initiated prematurely by enhancing GABA-mediated transmission with zolpidem (FIG. 4a), a GABA A α1-, α2- and α3-subunit selective ligand 46. Systematic use of the mouse knock-in mutation has further shown that only one of these subtypes, the α1-subunit-containing circuits, drives cortical plasticity (FIG. 4b). Diazepam fails to trigger premature plasticity in α1-subunit-mutants, although they are fully capable of undergoing plasticity at the proper age (postnatal day (P) 25) even without drugs 46, as they form normal α1- subunit-containing GABA receptors at the appropriate time. Importantly, GABA-releasing interneurons in the neocortex show precise connectivity 25,58 (FIG. 3b). Synapse formation by parvalbumin-positive cells is largely mediated through molecular cues and then refined by neuronal activity 65,66. These cells include axon-ensheathing CHANDELIER CELLS and soma-targeting LARGE BASKET CELLS. The latter extend the wide-reaching, horizontal axonal plexus (FIG. 1), which has been shown to span ocular dominance columns in the cat 67. The use of immuno-electron microscopy has further indicated that individual GABA A receptor α-subunits are trafficked to discrete postsynaptic sites on the pyramidal cell axon, soma and dendrites (FIG. 3b). For example, α2- subunits are preferentially enriched at the axon initial segment and at short-range basket cell synapses that are innervated by CHOLECYSTOKININ-positive axon terminals 68,69. Although α2-subunit-containing connections do regulate neuronal firing, they have no effect on the induction of plasticity 46. This dissociation has implications not only for models of brain development, but also for the safe design of benzodiazepines for use in human infants (see below). The α1-subunit-containing GABA A receptors are instead localized to receive parvalbumin-positive (but not cholecystokinin-positive) synapses on the soma 68,69, further implicating these large basket cell circuits in the control of the critical period. With age, large parvalbumin-positive cells are preferentially enwrapped in PERINEURONAL NETS of extracellular matrix (ECM) molecules and sugars 70. When these are disrupted, perisomatic inhibition of their targets is reduced 71, and ocular dominance shifts can once again be induced by monocular deprivation, even in adulthood 72. This might be the result of resetting and tapping the original GABA-mediated trigger 44 (FIG. 5b), as perineuronal nets probably control the extracellular ionic milieu (for example, potassium concentration 70 ) that surrounds parvalbumin-positive cells, allowing them to establish their fast firing efficiency 60 62, or might otherwise sequester molecular regulators of parvalbumin-positive cell maturation (S. Sugiyama et al., unpublished observations). The identification of a cellular critical period trigger holds great promise as a therapeutic target and for the development of strategies for lifelong learning. Expression by proteases: structural rewiring The ECM is increasingly being recognized as a potent site for critical period plasticity 73. To convert physiological events (altered vision) into structural refinements, connections must ultimately be broken and neuronal wiring rerouted. Proteases are ideally suited to clear the way for growing neurites 74. TISSUE-TYPE PLASMINOGEN ACTIVATOR (tpa) is the main serine protease in the postnatal mammalian brain 75,76, and was originally identified there as an IMMEDIATE EARLY GENE activated by hippocampal seizures 77. Proteolysis by tpa is gradually upregulated in V1 within 2 days of monocular deprivation during the critical period (FIG. 5a), but not in adults or Gad65-knockout mice 78. Interestingly, at least 2 days of diazepam treatment are required to rescue plasticity in the absence of GAD65 (REF. 45). Functional ocular dominance plasticity is impaired when tpa action is blocked and can be rescued by exogenous tpa (but not diazepam) 78. Permissive amounts of tpa might, therefore, couple functional to structural changes downstream of the excitatory inhibitory balance that triggers visual cortical plasticity. The second messenger systems, which are known to be recruited during ocular dominance 880 NOVEMBER 2005 VOLUME 6

5 PARVALBUMIN One of three calcium-binding proteins, which, together with calretinin and calbindin, are expressed in most GABAmediated neurons in the neocortex in largely nonoverlapping groups. FAST-SPIKING Ability of parvalbumin-positive cells to fire non-adapting action potentials at rates of up to several hundred Hertz, due, in part, to unique potassium conductances (Kv3 class). CHANDELIER CELL Stereotypical GABA cell of the cerebral cortex that ensheathes the axon initial segment of up to 200 pyramidal cells with cartridge synapses to directly control action potential generation. LARGE BASKET CELL Class of GABA cell with long, horizontally-extending axon that makes potent inhibitory contacts on the soma and proximal parts of the dendrites of target pyramidal cells. CHOLECYSTOKININ (CCK). A neuropeptide that is found in non-fast-spiking large basket cells that contact the soma of target pyramidal cells at synapses enriched in GABA A receptor α2-subunits. PERINEURONAL NET A conglomeration of chondroitin sulphate proteoglycans, extracellular matrix and cell-adhesion molecules that condenses around particular large-basket cells with age. TISSUE-TYPE PLASMINOGEN ACTIVATOR (tpa). The major serine protease in the brain, well known as an anti-clotting agent that works by cleaving fibrin in the blood. IMMEDIATE EARLY GENES Transcription factors that are induced within minutes of intense neuronal activity. Examples include zif268 and c-fos. PLASMIN The active form of plasminogen, which is the primary target of tpa action, and itself a protease that is known to cleave extracellular matrix molecules (for example, laminin and phosphacan). a GAT-1 GABA GABA A R β γ2 α Gephyrin GABA A receptor subunits Glu GAD65 π ρ θ β4 β3 β2 β1 ρ3 ρ2 ρ1 γ3 ε Cl Ca 2+ b GABA B Benzodiazepine γ2 γ1 α6 α4 α5 α2 α3 α1 plasticity 81 (FIG. 6), lie along a molecular cascade that links neuronal activity to tpa release 4,78, the structural consequences of which have recently been clarified. Thalamocortical axon rearrangement is a classical outcome of critical period plasticity (FIG. 1), but is much too slow to explain the rapid shift of ocular dominance that occurs within days of monocular deprivation 10 13,82. Morphological plasticity is initiated postsynaptically along the apical dendrites of target pyramidal cells in the cerebral cortex, where spines serve as pleiomorphic sites of excitatory synaptic connection. Spine shape has been shown to be highly dynamic using two-photon laser scanning microscopy in living transgenic mice expressing green fluorescent protein (GFP) in a subset of layer 5 cells. The motility of spines decreases with age in the visual cortex 83,84, but, during the critical period, can be transiently elevated by 2 days of monocular deprivation 85 (FIG. 5a, step 1). This occludes the motility that can be induced by the direct application of tpa to naive brain slices, which indicates that tpa and its substrate, plasminogen, might be the endogenous mediators of experience-dependent spine motility. Even along the same apical dendrite 85, spines Ch Large basket cell (CCK) 6 Grey matter White matter α2 α1 α3 α5 α2 α1 ais α2 DB Ng CRC Thalamocortical afferents Large basket cell (PV + ) Figure 3 Heterogeneity of local GABA circuits in the neocortex. a Inhibitory synaptic transmission is mediated by GABA (γ-aminobutyric acid) synthesized by glutamic acid decarboxylase 65 (GAD65) in the presynaptic terminal during strong stimulation. The postsynaptic effect of released GABA is modulated by exogenous benzodiazepines acting on specific postsynaptic receptors. Among the 20 identified subunits that the GABA A (GABA type A) receptor (GABA A R) may comprise 20,63, the α1-, α2-, α3- and α5-subunits contain an amino acid residue, histidine, that is vital to the benzodiazepine binding site 20,64. Benzodiazepine (but not GABA) binding is lost when this amino terminus histidine is mutated to arginine (causing the α1-, α2-, α3- and α5-subunits to mimic the naturally benzodiazepine-insensitive α4- and α6-subunits. b Many subtypes of GABAreleasing inhibitory interneuron can be identified in the neocortex on the basis of morphology, connectivity, expression of calcium-binding proteins or neuropeptide content 56 58,25. Moreover, specific contacts are preferentially enriched in specific GABA A receptor α-subunits 63,64. The perisomatic localization of different subunits is shown next to the central pyramidal neuron. All subunits are found diffusely along the dendrite. ais, axon initial segment; CCK, cholecystokinin expressing; Ch, chandelier cell; CRC, Cajal Retzius cell; DB, double bouquet cell; M, Martinotti neuron; N, neurogliaform neuron; PV +, parvalbumin positive. Lower part of panel a modified, with permission, from REF. 63 (2001) Elsevier Science. are set in motion by brief monocular deprivation only in layers 2, 3 and 5, consistent with early extragranular changes that instruct later events in layer 4 (REF. 12). After 2 days of monocular deprivation, increased proteolysis degrades ECM and cell-adhesion proteins before any physiological ocular dominance shift is detectable 38,78. Consequently, spines are eliminated by 4 days of monocular deprivation, which corresponds to the rapid, complete loss of functional responsiveness 86 (FIG. 5a, step 2). After this postsynaptic pruning, axons carrying input from the deprived eye retract before afferents serving the open eye can migrate to spaces cleared by tpa PLASMIN along the dendrite (FIG. 5a, step 3) 10,11. As expected for a true critical period event, spine density is not reduced by brief monocular deprivation in adulthood or in mice that lack tpa or GAD65 (REF. 86). The reduction in spine density can be restored in tpa- or Gad65-knockout mice by exogenous tpa or diazepam infusion, respectively. Importantly, spine motility and pruning faithfully reflect competitive interactions between the two eyes, as they fail to occur in the adjacent monocular segment that receives input solely from the contralateral eye 85,86. M NATURE REVIEWS NEUROSCIENCE VOLUME 6 NOVEMBER

6 a % of cells % of cells MD Critical period BDZ/veh P15 P Vehicle Zolpidem Ocular dominance LONG-TERM DEPRESSION (LTD). A persistent reduction of synaptic transmission in response to weak, poorlycorrelated input. LONG-TERM POTENTIATION (LTP). A persistent strengthening of synaptic transmission in response to strong, correlated input. b GABA A R knock-in % of cells % of cells α1(h101r) + diazepam at P α1(h101r) at P Ocular dominance Figure 4 Specific GABA A circuits for visual cortical plasticity. a Ocular dominance shifts, as rated by the neuronal discharge of single units from the mouse visual cortex on a scale from 1 (contra) to 7 (ipsi) eye input, can be induced by monocular deprivation (MD) before the critical period in the presence of zolpidem 46, a benzodiazepine (BDZ) agonist that is selective for α1-, α2- or α3-subunit-containing GABA A (γ-aminobutyric acid type A) receptors (GABA A Rs) 20. b Conversely, premature plasticity cannot be induced in mice with a point mutation that renders the α1-subunit insensitive to the benzodiazepine diazepam 46 (α1(h101r)). Importantly, GABA A receptor expression and localization occur normally, as the knock-in mice show robust plasticity (without drugs) at the typical critical period (postnatal day (P) 25). Figure modified, with permission, from REF. 46 (2001) Elsevier Science. Ultimately, the territory that represents the open eye grows 9 11 (FIG. 5a, step 3, right), and this requires cortical protein synthesis 87. Cleavage of secreted proneurotrophins by tpa plasmin to yield active mature forms such as BDNF might contribute to the elongation of neurites 88,89. As proteolytic activity slowly wanes after a week of monocular deprivation 78, axons serving the open eye expand 10,11, spines grow out to meet them and their density largely recovers 86 (FIG. 5a, asterisks). Curiously, spine loss is most robust near the soma of layer 2/3 pyramidal cells. Competition detected by inhibitory circuits impinging on pyramidal cell somata might, therefore, be translated into structural changes mediated by a multistep proteolytic action of the extracellular tpa plasmin cascade. The homosynaptic view It is tempting to speculate that the loss or gain of visual responsiveness of neurons in V1 during the critical period is simply the result of homosynaptic LONG-TERM DEPRESSION (LTP) or POTENTIATION (LTP) somewhere in the visual circuit (the homosynaptic view). On theoretical grounds, homosynaptic rules of excitatory synaptic plasticity alone are insufficient to produce a competitive outcome 90, as they require the involvement of other complex mechanisms, such as sliding thresholds or metaplasticity. Although correlative evidence has been presented to support homosynaptic rules for plasticity in vivo, the fact that direct comparisons are not supportive is often overlooked. In favour of the homosynaptic view, changes in phosphorylation state and membrane trafficking of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptor subunits events that seem to underlie LTP or LTD at central synapses 91 have been observed in the primary sensory cortex after natural sensory experience in vivo. For example, whisker stimulation at P12 14 in rats drives a recombinant AMPA receptor subunit (GluR1) into synapses from layers 4 to 2/3 in the somatosensory barrel cortex 92. Expression of the GluR1 cytoplasmic tail, which inhibits the synaptic delivery of endogenous receptors during LTP in vitro, blocks this insertion and subsequent synaptic potentiation in vivo. Whisker deprivation produces both an expansion of neighbouring input from spared barrels and a loss of response to the principal whisker. The latter largely occludes induction of NMDA (N-methyld-aspartate) receptor-dependent LTD by low-frequency stimulation (LFS) along the principal ascending pathway 93, which supports the idea that the mechanisms mediating changes in barrel cortex responsiveness are the same as those that mediate LTD. Similarly, in the visual cortex, 24-h monocular deprivation (too short a time period to shift ocular dominance in vivo 38 ) reduces the saturation level of LTD in response to repeated LFS, and leads to numerous changes in the GluR1 phosphorylation status, which are akin to those that occur during hippocampal LTD 94. The main difficulty with correlative studies is that they can never demonstrate causality. Ostensibly similar molecular pathways (FIG. 6) might be used differently by LTP/LTD and by critical period mechanisms subserving circuit function in vivo. For example, the late onset and experience-dependent profile of NMDA receptor 2A (NR2A) subunits known to determine LTP levels in the hippocampus 99,100 is irrelevant to critical period expression in the visual cortex 53. A decline in LTP/LTD magnitude and slow emergence of NR2A at thalamocortical synapses in the barrel cortex during the anatomical critical period in the first postnatal week might also be coincidental 101,102. Barrel shrinkage and expansion on whisker cauterization at birth is impervious to cortical NMDA receptor deletion 103 and might reflect changes at subcortical levels, before thalamic axons segregate in layer 4 (REF. 104). In the absence of NR2A, the depolarizing action of NMDA currents is prolonged. Like Gad65 deletion 43, this tips local circuit equilibrium towards excitation, weakening ocular dominance plasticity that can be restored by diazepam 53. Counter-intuitive to an LTP perspective 99,105, inhibition is required for plasticity in vivo when NMDA receptor function is high. Conversely, although LTD is blocked by NMDAreceptor antagonists, high doses of APV 35 promote a paradoxical loss of open eye input, as does strong postsynaptic activation of GABA A receptors with muscimol 36,37. Although maturation of other receptive field 882 NOVEMBER 2005 VOLUME 6

7 a 1 Detection 2 d MD Spine motility (tpa plasmin) No OD shift 2 Expression 4 d MD Spine elimination Axon retraction Full OD shift 3 Consolidation 15 d MD Spine recovery Axon outgrowth consolidated b * * * SPIKE-TIMING DEPENDENT PLASTICITY Describes physiological windows for LTP or LTD of synaptic transmission based on the arrival time of incoming action potentials with respect to back-propagated spikes in the target dendrite. * Large PV + basket cells PNNs Motility Spine GABA Glu tpa plasmin pro BDNF Pruning Axon Sprouting Mature BDNF MD Days Functional plasticity (rapid) Structural rewiring (long-lasting) Detection Expression Consolidation Figure 5 Structural consolidation during the critical period. a Once functional detection of competing inputs is made possible by the maturation of relevant GABA (γ-aminobutyric acid) circuits 43 48, a series of structural rearrangements accompanies sensory deprivation. Shortly (2 days) after monocular deprivation (MD), the motility of spines (two-headed arrows) 85 is increased on apical dendrites of excitatory pyramidal neurons by an increase in tissue-type plasminogen activator (tpa) plasmin proteolytic activity (blue background) 78. No shift in ocular dominance (OD) can be detected at this stage 38 (1). The total number of spines on pyramidal cell apical dendrites is transiently and significantly decreased owing to their elimination (asterisks), with a time course that corresponds to the loss of cellular responses of the deprived eye 4 days after monocular deprivation 86. This is followed by a retraction of thalamocortical axons 10,11 as extracellular tpa plasmin activity remains high 78 (2). After long-term monocular deprivation, new spines emerge (asterisks) to receive synaptic input from sprouting axons carrying input from the open eye (triangles) 10,11,13 as tpa plasmin activity subsides 78 (3). b These results predict that structural modification is necessary to reactivate critical period plasticity in adulthood. Infusion of chondroitinases (scissor symbols) to break up perineuronal nets (PNNs) in the extracellular matrix restores ocular dominance shifts to adult rats 72. Interestingly, these nets preferentially enwrap large, parvalbumin-positive (PV + ) basket cells 70, which control perisomatic inhibition 71 and trigger the endogenous critical period. The section depicts parvalbumin-positive basket cells stained with monoclonal antibody (red) and perineuronal nets stained with Wisteria floribunda agglutin (WFA; green). Note that small parvalbumin-positive cells are not ensheathed by perineuronal nets. BDNF, brain-derived neurotrophic factor; Glu, glutamate. properties (such as orientation bias 44,53 ) might reflect subunit-specific coupling of NR2A to postsynaptic signalling pathways (like LTP 100 ), the yin-yang relationship of excitatory inhibitory balance is essential for ocular dominance plasticity. During learning, distinct plasticity mechanisms subserve sequential phases of memory 106. Local (synaptic and cellular) events are eventually consolidated at a systems (circuit) level by increasing protein synthesis and neurite growth. Excessive emphasis on LTP/LTD alone (which is all-or-none at single synapses 107 ) ignores the gradual, long-lasting changes that are the hallmark of critical period plasticity. One day of monocular deprivation does not occlude homosynaptic depression by single LFS 94, and these early forms of synaptic change, which persist in the presence of protein synthesis inhibitors 108, are insufficient to shift ocular dominance in vivo 87. However, multiple spaced conditioning stimuli, used to saturate LTP/LTD 94, activate distinct transcriptional pathways (for example, that of immediate early gene Zif268) 109, which are also not required for ocular dominance plasticity 110. In fact, the role of proteases (tpa) in growth for long-lasting forms of LTP 111,112 is opposite to that for ocular dominance plasticity (spine pruning 86 ). Overall, no consistent correlation has been found between the ability to induce homosynaptic plasticity in vitro and amblyopic effects in vivo The relationship is not straightforward, as the mechanisms of LTP and LTD might differ depending on cortical layer 117. Moreover, unlike the very rapid induction of synaptic changes, no loss of visual response or acuity occurs until several days of eyelid suture have elapsed 38,39. Rapid phosphorylation dynamics of one GluR1 amino acid residue alone are unlikely to explain the complex functional and structural events that constitute the critical period (FIG. 5). Spine shrinkage by LFS is, in fact, mediated by cofilin and is independent of the AMPA receptor dephosphorylation that underlies LTD 118. At present, there is only limited evidence directly connecting the coincident structural and functional plasticity at spines 119. Most strikingly, synaptic depression is intact whereas ocular dominance plasticity is lost in mice that conditionally overexpress the protein phosphatase, calcineurin 120. Underscoring a double dissociation, homosynaptic models based on NMDA-receptor activation stipulate that the maturation of inhibition will terminate plasticity 105, whereas quite the opposite is true: in vivo, GABA function is required to trigger the critical period 44. Given that diazepam 121 and endogenous BDNF 122 block LTD induction in the cortex, their acceleration of plasticity in vivo is not predicted by homosynaptic models, which are routinely studied in vitro in the presence of GABA-receptor blockers. In the end, adjusting local inhibitory circuit function has provided direct control over the critical period (FIG. 2), which LTP-based models have not. Local circuit models of critical period plasticity That inhibition might enable plasticity in the developing brain seems paradoxical. The identification of a particular GABA circuit that drives the onset of the critical period and the subsequent sequence of anatomical events (FIG. 5) now allows us to construct realistic models with which to target the plasticity process more precisely. Two scenarios centred on the parvalbumin-positive basket cell serve as heuristic examples for further study. One is an instructive model (FIG. 7a), in which powerful, fast somatic inhibition edits one-by-one the action potentials that can pass into the dendritic arbor by back-propagation through the cell body. Recent SPIKE-TIMING DEPENDENT models of synaptic plasticity rely on precise millisecond NATURE REVIEWS NEUROSCIENCE VOLUME 6 NOVEMBER

8 BDNF Neuromodulators Plasminogen Plasmin Sensory experience GABA/glutamate PKA/CAMKII/calcineurin ERK/CREB PROTEIN KINASE A (PKA). Phosphorylates multiple targets (including AMPA and GABA A receptors) when cyclic AMP binds its regulatory subunits to release the catalytic domains. tpa Neurite motility Myelination BDNF PAI-1 Mature Pro Functional plasticity (rapid) Critical period Structural rewiring (long-lasting) Detection Expression Consolidation Figure 6 Molecular mechanisms of visual cortical plasticity. Many candidate plasticity factors have been screened using the monocular deprivation protocol by pharmacology in kittens or by gene-targeted disruption in mice (for a review, see REF. 81). Only a handful of second messenger molecules have been found to have a direct role in plasticity without perturbing global neuronal activity. These include PROTEIN KINASE A (PKA) 181, calcium/ calmodulin-dependent protein kinase II (CaMKII), calcineurin 120, extracellular signal-regulated kinase (ERK), cyclic AMP responsive element binding protein (CREB), protein synthesis machinery 87 and the plasmin system (tissue-type plasminogen activator (tpa) plasmin), which is regulated by its inhibitors (plaminogen activator inhibitor, PAI-1) Brain-derived neurotrophic factor (BDNF) has an early role in plasticity, establishing the GABA (γ-aminobutyric acid) cells that discriminate competing sensory inputs to trigger the critical period 47. Mature BDNF, which is produced from the cleavage of pro-bdnf by tpa plasmin 88, in turn stimulates the expression and release of tpa 185. Both tpa and BDNF can then contribute sequentially to the final anatomical rewiring of the cortical circuit 10,11,13,85,86,89 (FIG. 5). Plasticity might come to an end when permissive factors are gradually lost 86, or when further growth is actively suppressed by late-emerging inhibitory factors (myelin) in the extracellular matrix 7,73,182. time windows to allow such postsynaptic spikes to meet presynaptic input 123,124. Sloppy gating by weak inhibition at the soma would prevent a competitive outcome by allowing excess back-propagation and spurious coincident activity with infrequent, deprived inputs from the retina 125. Identifying a molecular substrate that is unique to spike-timing dependent (as opposed to LFS or tetanusinduced 126 ) plasticity will be a prerequisite to testing this scheme in vivo. Both immature (pre-critical period) and Gad65- knockout mice show prolonged neuronal discharge that continues well after stimuli have passed through the receptive field of individual cells Chandelier cell cartridge synapses (which use α2 GABA A receptor subunits) can directly control this excessive spiking at the axon initial segment 46,127. Surprisingly, plasticity is still triggered prematurely by diazepam when hyperexcitability persists in mice whose α2-subunits can no longer bind benzodiazepines 46. Fast-spiking, feedforward inhibition mediated by α1-subunit-containing receptors on the soma is then ideally situated to suppress back-propagation of unwanted spikes 125 (FIG. 7a). The model incorporates the wide-reaching, horizontal axons of basket cells that receive input from one eye to inhibit targets of the other eye 67. This simple contrast enhancement circuit relieves the burden of discriminating competitors by LTP/LTD at single synapses. Alternatively, well-timed spikes in the soma itself might facilitate nuclear calcium entry for gene transcription (and growth) in response to waves of calcium arriving from synapses on dendritic spines. A second, permissive model emphasizes newfound knowledge of parvalbumin-positive cell biology. Even in adulthood, basket cells can be coupled electrically into groups of 40 or 50 cells 128,129, endowing the network with the ability to detect synchrony 130 (FIG. 7b). Whereas simultaneous inputs (for example, from the same eye) rapidly co-excite cells through gap junctions, even a 2-ms input jitter (for example, between opposite eyes) is sufficient to dampen the coupling by reciprocal GABA A synapses, which are also enriched in α1-subunits 131. As a result, these neurons with long, horizontal axonal arbors 67, are maximally active on a columnar scale, timelocked to release growth or plasticity factors when strong synchronous activity arrives in the neocortex. Gap junctional coupling between interneurons also promotes synchronous oscillation among principal cells 132,133. Validation of the permissive model awaits conditional deletion of connexins in cortical parvalbumin-positive cells alone, as the retina is also rich in gap junctions 134. This model assumes an extracellular locus of competition to be quite distinct from the intracellular mechanisms of LTP and LTD. Axons and dendritic spines that are co-active with the synchronized parvalbuminpositive cell network can benefit from the rapid, focal creation of a permissive growth environment 78,86,135. The source and dynamics of tpa plasmin release in the brain remain unclear owing to the lack of specific reagents. Laminar motility of spines 85 and their rapid pruning 86 by brief periods of monocular deprivation could reflect activity-dependent secretion of proteases from the axons of fast-spiking cells themselves, in which parvalbumin is an important contributor to presynaptic calcium signals and synaptic integration 139. This might explain why spines nearest the soma of layer 2/3 pyramidal cells are most robustly lost during monocular deprivation 86, as they lie nearest the parvalbuminpositive cell-rich layer 59. Potential synaptic plasticity of GABA-mediated transmission might add a further layer of regulation. How a competitive outcome arises by uniformly bathing dendrites in proteases also needs to be considered. Cell-adhesion molecules can become insensitive to proteases during high levels of activity 140,141. Less active synapses also release fewer endogenous protease inhibitors 142, tilting the overall balance nearby towards pruning. Therefore, open-eye axons can protect themselves from proteolytic removal in two ways, whereas silent deprived-eye inputs are left defenceless in a sea of tpa plasmin during monocular deprivation. Towards a common brain principle How general is the idea of local circuit inhibition and critical periods across other systems? Zebra finches acquire a unique song once in life during a critical period for vocal learning that progresses through three stages NOVEMBER 2005 VOLUME 6

9 a Instructive b Permissive Activity-dependent plasticity factors LTP LTD LTP LTD Input Spike backpropagation Axonal inhibition K v 3.1 PV + α2 α1 Somatic inhibition PV + K v 3.1 Network output Coupled PV + cells Excitatory input Synchronous (same eye) α1 Non-synchronous (opposite eyes) Figure 7 Two models for inhibitory control of sensory plasticity. a Somatic inhibition by fast-spiking parvalbumin-positive (PV + ) cells (green) expressing the potassium channel Kv3.1 is mediated by the GABA A (γ-aminobutyric acid type A) α1-subunit, whereas axonal inhibition is mediated by the GABA A α2-subunit. Somatic inhibition is ideally situated for suppression, or editing of unwanted spikes, preventing them from back-propagating through the cell body into the dendritic tree (red arrow). This is an instructive model, as individual action potentials can produce long-term potentiation (LTP) or depression (LTD) based on precise spike-timing dependent windows at individual synapses of coincident pre- and postsynaptic activity 123,124. Notably, failure to regulate excess spiking at the axon initial segment can still be differentiated by fast-spiking inhibition at the cell body. b Gap junctional coupling endows networks of parvalbumin-positive (PV + ) interneurons with the ability to detect synchronous input Even a slight jitter in input timing (for example, between eyes) dampens network activity through reciprocal GABAmediated contacts (enriched with GABA A -receptor α1-subunits 131 ). Only synchronous open-eye input will produce maximal, activity-dependent release or uptake of permissive factors for neurite growth (for example, tissue-type plasminogen activator 78,86 and brain-derived neurotrophic factor). Competition is determined extracellularly. After first memorizing the father s template, birds enter a motor practice phase, followed by song crystallization. In register with the active singing component, a peak in the number of GABA neurons is observed in the motor nucleus robustus archistrialis (RA), where descending auditory and forebrain memory circuits converge onto individual cells 143. Such a GABA profile is observed only in males, and not in females, which do not produce song. Local inhibitory circuits might allow the proper matching (and discrimination) of template and vocalized sub-song in the RA before consolidation. Indeed, infusion of cannabinoids, which primarily bind presynaptic CB1 receptors to reduce GABA release 144, alters developmental sensorimotor vocal learning but not adult song performance 145. If barn owls are reared wearing prism lenses to skew their visual world, a learned shift occurs in the auditory map of interaural time differences in the tectum so that it corresponds with the visual input 146. Multiple maps can be stored in the same tectum if this experience occurs during a critical period 147. To allow discrimination between conflicting circuits, new GABA-mediated connections are formed to suppress the unused representation 148. Adult plasticity is limited to incremental displacement of the visual scene within the architectural constraints of axons that were hardwired during the critical period 149,150. Conversely, in the barrel cortex GABA circuits are formed and reorganized throughout life , and are associated with lifelong plasticity 154. This is also the case in the mammalian olfactory system, where constant neurogenesis is responsible for the underlying memory of odour discrimination in adulthood 155,156. The newly born cells are GABA-containing granule cells, the functions of which include lateral inhibition and synchronization of neuronal activity 157,158. This raises the question of whether an olfactory critical period would emerge in the absence of neurogenesis, or whether visual plasticity could be maintained by prolonging cell proliferation in the neocortex. Most interestingly, the control of visual cortical plasticity by GABA in animal models might also apply to the development of the human brain. In postnatal samples of V1 that were removed post-mortem 159, the maturation of NR2A to 2B is complete within the first 9 months after birth. By contrast, Gad65 expression and the conversion of GABA A receptors from α3- to α1-subunit containing occurs more slowly, over several years, which is consistent with the extended length of the critical period for amblyopia in humans 6,7. Strikingly, the levels of GAD67 and α2-subunits are constant during the same early postnatal time period 159, consistent with their limited role in plasticity in animals 46. To avoid the rapid, premature induction of critical period plasticity through α1-subunit-containing receptors 45,46, new benzodiazepine agonists that are selective for α2- subunits should be developed for the control of status epilepticus in human infants 160. NATURE REVIEWS NEUROSCIENCE VOLUME 6 NOVEMBER